Welcome to NVIDIA's internal version of the renowned GPU Hackathons. This is an opportunity for engineers and developers to learn AI, with mentors and hands-on training, and compete to develop a deployable Deep learning pipeline using NVIDIA AI technology stack, for COVID-19 mass surveillance.
Participation in this event is limited to NVIDIA India employees only and on a first come first serve basis--we will accept a maximum of 40 teams who can participate in this event. For inquiries, please join the NVIDIA Slack channel at #gpu_ai_india_hackathon.
Whether you are a beginner or an ML expert, the GPU AI Hackathon is designed to provide you with all the tools you need to learn and apply knowledge of AI and Nvidia products for real life applications. The selected participants will be provided with the following during the Hackathon:
- Compute Infrastructure: Access to 1 NVIDIA Tesla V100 GPU powered system.
- Training: Participants will be provided training on latest HPC-AI tools including basics of Deep Learning, NVIDIA DeepStream SDK and AI GPU libraries from NVIDIA (wherever applicable).
- Mentorship: Mentors from NVIDIA DeepStream product and SA team will guide the participants to help achieve optimal results during the course of Hackathon.
The Problem Statement: Estimate Social Distancing using Intelligent Video Analytics (IVA):
The objective of the competition portion of the AI Hackathon is to find patterns on input sources like camera, satellite, social platforms which can help identify suspicious activities to stop or track the spread of COVID-19. Two persons are considered “too close” if the bounding boxes overlap over a certain threshold. This ensures that the metric is scale invariant (persons farther from the camera are treated the same as the ones closer). Further it’s required to create an application for this in NVIDIA Deepstream SDK in order to perform large scale deployment and analytics:
- If the distance between two people is less than a certain threshold, it should send an alert.
- If there are dense crowds in any region it should send an alert with location.
- Develop an efficient Neural Network to estimate social distancing. Beneficiary identification.
- Generate real time alerts via Deepstream SDK for mentioned anomaly.
- Performance in terms of number of input feeds with FPS (Frame Per Second)
- Accuracy with maximum true positive or recall.
Participants will submit the log details of FPS and Accuracy before the final demo.
- GitHub: Smart Social Distancing
- Jetson Project of the Month: Smart Social Distancing with AI
- Blog: Building Intelligent Video Analytics Apps Using NVIDIA DeepStream 5.0 (Updated for GA)
Sample Input Feeds:
- Video: Smart distancing with the help of AI - to control social distancing with AI
- Video: Walking in New York City during Coronavirus Pandemic
- Technological Superiority: Use of latest AI models, better performance and accuracy.
- Innovation Quotient: Innovative with research quotient and potential to become a unique solution.
- Viability: The solution is practically feasible in a real world scenario considering the cost to solution, ease of integration into existing infrastructure and easy deployment in the field.
Important Event Dates
NVIDIA India AI Hackathon: Questions Clarified
KGMON Invited Talk: How to Win ML Competitions
NVIDIA India AI Hackathon Application Deadline
NVIDIA India AI Hackathon: GPU Bootcamp Day 1
NVIDIA India AI Hackathon: GPU Bootcamp Day 2
NVIDIA India AI Hackathon Begins Online
NVIDIA India AI Hackathon Concludes
NVIDIA India AI Hackathon Final Solution Submissions and Presentations
Announcement of Winners